Multi-stage gender classification using BGLAM and eigenfaces

Abstract

In this thesis, we propose a novel approach for gender classification by two-level classifiers. The basic gray-level aura matrix (BGLAM) is used to extract texture feature from face component regions such as two eyes, nose, and mouth. The feature vectors are classified into male, female, or unknown using support vector machine (SVM) with the radial basis function (RBF) kernel. When the classification results are unknown, further classification is conducted. First, a whole head region is detected based on the location of face. The eigenfaces for the head region are obtained and the weight vector is estimated. The texture feature vector obtained using BGLAM and the structure feature obtained using eigenfaces are linearly combined using weights. Finally, the combined feature vectors are classified as male or female using SVM. The training samples are collected from FERET databases. In the experiment, the error rate of classification using BGLAM is compared with that of general local binary pattern (LBP). The proposed method is compared with traditional neural-network based method and local binary pattern based method.